To name a fallacy is to disarm it. The intellectual history of the last two and a half millennia, from Aristotle's Sophistical Refutations to the modern critical-thinking textbook, can be read as a slow, patient cataloguing of the ways in which reasoning goes wrong — the recurring traps into which clever minds, no less than careless ones, are prone to fall. The catalogue is not closed; new variants appear with every new medium, every new political moment, every new advertising campaign. But the underlying patterns are surprisingly stable.
A fallacy is not merely a false belief. A false belief may be sincerely arrived at by valid reasoning from mistaken premises; that is intellectual misfortune, not error in reasoning. A fallacy is a defect in the inference itself — a place where the conclusion fails to follow from the premises, where the premises smuggle in what they are supposed to prove, where the language used disguises a shift in meaning, or where the appeal moves the audience for reasons that have nothing to do with truth.
What follows is a working taxonomy of the most consequential fallacies, organized by the dimension of argument in which each defect lives: form, relevance, presumption, language, causation, and induction. For each, you will find a precise definition, the mechanism of its failure, and three concrete examples drawn from politics, science, commerce, and ordinary life. Read it slowly. Read it twice. The pleasures of clear thinking are durable, and the cost of muddled thinking — in personal decisions, in public deliberation, in our private peace — is steeper than we usually admit.
Part I
Errors of Form
When the very shape of the inference is invalid
A formal fallacy is a defect that can be identified without reference to the content of the argument at all. The logical scaffolding itself is broken; the conclusion does not follow from the premises by any rule of valid inference, no matter how plausible the subject matter. These are the cleanest, most diagnosable errors, because once the form is exposed, the failure is plain.
1
Affirming the Consequent
Modus ponens, deformed
Form: If P then Q. Q is true. Therefore P. — The reasoner mistakes a sufficient condition for a necessary one. Many causes may produce the same effect; observing the effect does not pin down which cause.
i.
Everyday"If it rains, the street is wet. The street is wet, so it must have rained." — But a burst water main, a street sweeper, or a child with a hose will leave the same wet street.
ii.
Medicine"Pneumonia causes coughing. The patient is coughing. Therefore she has pneumonia." — A hundred conditions produce a cough; the inference is reckless without ruling them out.
iii.
Business"Companies with strong cultures outperform the market. We outperformed the market last quarter, so our culture must be strong." — A lucky product cycle, a falling competitor, or sectoral tailwinds can produce the same result.
2
Denying the Antecedent
A variation of Modus tollens
Form: If P then Q. P is not true. Therefore Q is not true. — The reasoner forgets that the conditional only rules out the case "P true and Q false"; as for what happens when P is false, it says nothing.
i.
Law "If an organization is a legal person (P), then it has civil subject qualification (Q). This organization is not a legal person (¬P) (e.g., a sole proprietorship). Therefore, it does not have civil subject qualification (¬Q)." Although a sole proprietorship is not a legal person, the Civil Code grants it the status of "unincorporated organization," which can participate in lawsuits and sign contracts in its own name, so Q still holds.
ii.
Education "If a student is a registered student of Peking University (P), then they are a Chinese citizen student (Q). This student is not a registered student of Peking University (¬P) (e.g., a registered student of Tsinghua University). Therefore, they are not a Chinese citizen student (¬Q)." Tsinghua registered students are obviously Chinese citizens (and the academic records are interconnected), so denying the antecedent "Peking University" cannot negate the consequent "Chinese citizen."
iii.
Macroeconomics "If a country is a member of the G7 (Group of Seven) (P), then it is a high‑income developed country (Q). This country is not a member of the G7 (¬P) (e.g., the Republic of Korea). Therefore, it is not a high‑income developed country (¬Q)." Although South Korea is not a G7 member, both the IMF and the World Bank clearly classify it as a developed country, so Q does not depend on P at all.
3
The Undistributed Middle
A failure of categorical syllogism
Form: All A are C. All B are C. Therefore all A are B. — The middle term C, which is supposed to link A and B, is never taken in its full extension in either premise. Two things sharing a common category are not thereby identical.
i.
Schoolroom"All cats are mammals. All dogs are mammals. So all dogs are cats." — The absurdity is the lesson: shared category does not mean shared identity.
ii.
Bigotry"Terrorists are Muslims. Ahmed is a Muslim. So Ahmed is a terrorist." — Even granting (falsely) the first premise, no one is identified by their category alone.
iii.
Pseudoscience"Successful entrepreneurs are risk-takers. Gamblers are risk-takers. So gamblers will be successful entrepreneurs." — A shared trait does not transfer outcomes.
4
Affirming a Disjunct
When "or" is read as exclusive without warrant
Form: P or Q. P is true. Therefore not Q. — In ordinary logic, "or" is inclusive: both may be true. To affirm one and conclude the other is false treats "or" as if it forced a choice.
i.
Romance"Either he loves her for her looks or for her mind. He clearly admires her mind, so it must not be about looks." — The two are perfectly compatible; he may admire both.
ii.
Workplace"He's either incompetent or lazy. Reviewing his record, he's incompetent. So at least we know he isn't lazy." — Nothing prevents him from being both.
iii.
Political"The policy is either ideological or corrupt. It's plainly ideological — so the corruption charge falls away." — Policies can be ideological and corruptly implemented at once.
❦ ❦ ❦
Part II
Errors of Relevance
When the premises, however true, are simply beside the point
These fallacies retain the surface appearance of argument but sever the connection between what is said and what is claimed. The premises may be entirely accurate; they simply do not bear on the conclusion. The reasoner — or the audience — is moved by something other than logical force: by character, by feeling, by force of numbers, by force of authority, by sheer distraction.
5
Argumentum ad Hominem
"To the man" — attacking the arguer in place of the argument
The argument's content is left untouched while the speaker is impugned — their character, motives, history, or social standing. Whether the speaker is a saint or a scoundrel is independent of whether what they say is true. There are three principal sub-types worth distinguishing: the abusive, the circumstantial (attacking motive or self-interest), and the tu quoque ("you too").
i.
Abusive"You can't trust her climate analysis — she's a known narcissist." — Her psychological profile is irrelevant to the temperature data.
ii.
Circumstantial"Of course the oncologist recommends the drug; her department gets pharmaceutical funding." — The conflict-of-interest is worth noting, but it does not refute the medical evidence.
iii.
Tu Quoque"My father told me to stop smoking, but he smoked for forty years — so his advice is worthless." — Hypocrisy in the messenger does not falsify the message.
6
Guilt by Association
Damnatio per consortium
A claim is discredited because it is also held, or was once held, by some disreputable group. The reasoner exploits the audience's existing distaste to taint a proposition that must, in fairness, be evaluated on its own merits. Its most reflexive modern form is the reductio ad Hitlerum.
i.
Public health"Vegetarianism? Hitler was a vegetarian." — True and irrelevant to whether vegetarianism is healthy or ethical.
ii.
Policy"Universal healthcare? That's what the Soviets had." — Whatever overlap exists, the merits of the policy must be examined directly.
iii.
Science"His paper cites a journal that once published a now-retracted study, so we shouldn't take him seriously." — Citation traces are no substitute for examining the paper's own argument.
7
The Red Herring
A diversionary scent across the trail
A topic adjacent to, but distinct from, the issue at hand is introduced — usually one easier to defend, or more emotionally arresting — in the hope that the audience will pursue the new scent and forget the original question. The technique is the workhorse of evasive politicians and uncomfortable spouses alike.
i.
Politics"Senator, why did your office overspend its budget?" — "What's truly outrageous is how little we spend on veterans in this country." — A worthy topic, deployed to escape the asked one.
ii.
Domestic"You forgot our anniversary." — "Did you see how much work I put into the garage this weekend?" — The garage labor is real; the question of the anniversary remains unanswered.
iii.
Corporate"The product has a safety defect." — "Let me remind you of our company's forty-year history of philanthropy." — A noble history does not unsharpen the present blade.
8
Appeal to Popularity
Argumentum ad populum — the bandwagon
A proposition is held to be true because many people believe it, or fashionable because many people endorse it. Yet popular belief and truth are independent quantities. The earth's flatness, the geocentric heavens, and the medical use of bloodletting were each, in their day, the consensus of the populace and the learned.
i.
Marketing"Over four million units sold worldwide — you can't argue with the numbers." — You can: four million people may have been mistaken in the same way.
ii.
Investment"Everyone is buying this stock; it must be a winner." — The most consistent feature of bubbles, from tulips to tokens, is unanimous enthusiasm at the peak.
iii.
Cultural"Astrology is wildly popular among educated people now — there must be something to it." — A revival of belief is not a revival of evidence.
9
Appeal to Improper Authority
Argumentum ad verecundiam misapplied
Citing an authority is not in itself fallacious — expertise is a legitimate input to belief. The fallacy occurs when the authority cited is operating outside their domain of competence, or is anonymous, or is dissenting against a well-established consensus without addressing the evidence behind that consensus.
i.
Celebrity endorsement"This actor's wellness brand is endorsed by an Oscar winner." — Acting talent is no qualification for biochemistry.
ii.
Cross-domain"A Nobel laureate in physics says this vitamin cures the common cold." — The prize is in the wrong field; the laureate may simply be wrong.
iii.
Vague authority"Studies show that..." — Which studies? By whom? In what journal? An invoked but unnamed authority is no authority at all.
10
Appeal to Emotion
Argumentum ad passiones
Strong emotion is summoned — fear, pity, outrage, disgust, pride — in lieu of evidence, so that the audience's assent flows from feeling rather than judgment. Emotions are not, in themselves, illegitimate in argument; the fallacy lies in substituting them for the work that reasons must do.
i.
Fear (ad metum)"If we don't pass this surveillance law, your children will not be safe in their schools." — The graphic image displaces a careful weighing of costs and effectiveness.
ii.
Pity (ad misericordiam)"I know I missed the deadline, but my grandmother is ill and I've barely slept." — The hardship may be real; whether the work merits a passing grade is a separate question.
iii.
Outrage (ad indignationem)"Are we really going to let them get away with this? Are you with us or with them?" — The provocation foreshortens deliberation; one is being herded, not argued with.
11
Appeal to Force
Argumentum ad baculum — "to the cudgel"
Acceptance of a claim is secured not by reasons but by threat — of violence, ostracism, professional damage, or social consequence. The threat may be implicit and polite, but it remains a substitute for evidence.
i.
Workplace"I think you'll agree the report's conclusions are sound — given that your performance review is next month." — Assent purchased by fear is not assent to truth.
ii.
Diplomacy"You should reconsider your trade policy; we have a great many ships in your waters." — A coherent argument might exist behind the gunboat, but the gunboat is not it.
iii.
Social"Anyone who questions this orthodoxy will be reported and removed from the group." — Conformity by intimidation is not consensus by reason.
12
Appeal to Nature
The naturalistic fallacy
Whatever is "natural" is held to be good, healthy, or right; whatever is "artificial" or "synthetic" is held to be suspect. The inference is invalid in both directions. Hemlock and arsenic are natural; insulin and antibiotics are synthetic. Moral or practical merit must be argued independently of origin.
i.
Wellness"This supplement is all-natural — so it can't hurt you." — Many of the most lethal toxins in pharmacology are derived from plants.
ii.
Ethics"It's natural for the strong to dominate the weak — therefore it's how society should be ordered." — The fact-value inference is unauthorized; nature is not a normative authority.
iii.
Diet"Humans weren't designed to drink milk after infancy; therefore dairy is unhealthy." — "Designed" smuggles in teleology; "unhealthy" requires evidence the design-talk cannot supply.
13
The Genetic Fallacy
Judging the idea by its origins
An idea is accepted or rejected based on where it came from — who first proposed it, the era, the institution, the prevailing culture — rather than on what it actually claims and how well it is supported. Provenance is one consideration among many; it is rarely a verdict.
i.
Historical"The structure of DNA was first guessed at by a man with appalling personal views — so the science is tainted." — The science is testable independent of the scientist.
ii.
Cultural"Yoga comes from a religious tradition I do not share, so its physiological benefits cannot be real." — Origin and effect are unconnected.
iii.
Politics"That economic proposal was first floated by the other party — so it must be bad." — Tribal sourcing replaces argument with allegiance.
❦ ❦ ❦
Part III
Errors of Presumption
When the argument quietly assumes what it is meant to prove
These fallacies are the most insidious, because the smuggled premise is rarely visible on the surface. The argument looks like it is doing work; in fact it is travelling in a tight circle, or resting on a question never posed, or appealing to absence of evidence as if it were evidence of absence. Catching them requires the willingness to stop, slow down, and ask: what would I have to already believe for this to seem compelling?
14
Begging the Question
Petitio principii — assuming the conclusion in the premises
The conclusion the arguer is trying to prove is already contained, often only slightly rephrased, in one of the premises. Anyone who does not yet accept the conclusion has no reason to accept that premise either; the argument therefore persuades only the already persuaded.
i.
Theology"The scripture is reliable because the scripture says it is the inspired word of God." — The premise borrows the conclusion's authority.
ii.
Politics"Free speech for hateful views must be restricted, because nobody should be permitted to speak hatefully." — Whether they should be permitted is the question.
iii.
Investment"This asset is undervalued because the market hasn't yet recognized its true worth." — "True worth" simply is "the value the market has not yet recognized" — the claim is sealed against test.
15
False Dilemma
The false dichotomy — a choice between two when many exist
The argument restricts the field of options to two, when in fact a continuum or richer landscape of alternatives is available. Usually one of the two is made grotesque so that the other — the speaker's preferred option — appears the only sensible choice.
i.
Wartime rhetoric"You're either with us, or you're with the terrorists." — The space between unconditional support and complicity is vast and unmentioned.
ii.
Parenting"Either we let her stay out as late as she wants or we lock her in her room." — Curfews, conversations, and trust exist on a wide spectrum between these extremes.
iii.
Technology"Either we adopt this AI tool company-wide today or we'll be irrelevant in five years." — Pilots, partial rollouts, alternative vendors, and waiting are all options the framing erases.
16
Argument from Ignorance
Argumentum ad ignorantiam — absence of evidence as evidence of absence (and the converse)
A claim is held true because it has not been proven false, or held false because it has not been proven true. In most domains, especially those where searches are incomplete, neither direction warrants the conclusion. The default of "we don't know" is often the honest one.
i.
Cryptozoology"No one has ever conclusively proven that the Loch Ness Monster doesn't exist — so it must exist." — The absence of disproof is not proof; the burden lies with the claim.
ii.
Politics"There's no evidence that the senator took the bribe — so plainly she didn't." — Investigations may simply have failed to find evidence that exists.
iii.
Medical"No study has ever shown this herb is harmful, so it must be safe." — Untested is not the same as tested-and-found-safe.
17
Complex (Loaded) Question
Plurium interrogationum — the question that hides an answer
A question is posed that cannot be answered — yes or no — without conceding a controversial assumption embedded in the question itself. The trap is that any direct answer ratifies the hidden premise.
i.
Classical"Have you stopped beating your wife?" — Yes or no, the respondent affirms a history of beating.
ii.
Journalism"Why did your administration ignore the warnings?" — The premise that warnings were ignored is the very thing in dispute.
iii.
Survey design"How concerned are you about your representative's record of corruption?" — "Corruption" is presupposed; the survey extracts only the degree of concern.
18
No True Scotsman
The retreat into ad hoc redefinition
When a counter-example threatens a universal claim, the arguer modifies the category to exclude the counter-example — not on principled grounds, but precisely to save the claim. The pattern: assertion → counterexample → "ah, but no true X would do that."
i.
The original"No Scotsman puts sugar on his porridge." — "My uncle is a Scotsman and he does." — "Then he's no true Scotsman."
ii.
Politics"Real conservatives never support tax increases." — "But Senator X is a lifelong conservative and supported one." — "Then he was never a real conservative."
iii.
Religion"No true Christian commits violence in the faith's name." — "But historically, many have." — "Then they weren't really Christians."
19
Special Pleading
Requesting an exemption without principled grounds
A general rule is applied to others but suspended for oneself or one's preferred case, with no defensible reason for the exception. The exemption is invented after the fact, tailored precisely to the awkwardness it must resolve.
i.
Personal"I know I said no phones at dinner, but mine's for work tonight — that's different." — The rule applies until it inconveniences the rule-maker.
ii.
Pseudoscience"My psychic abilities only work when no skeptics are present — their negative energy interferes." — The exemption conveniently insulates the claim from the only conditions under which it could be tested.
iii.
Corporate"Yes, we apply ethical standards to all suppliers — except this one, which is strategically important." — Importance is not a moral exemption; the principle dissolves on contact with interest.
20
Self-Sealing Arguments
The unfalsifiable position
A theory is so constructed — or so re-constructed in the face of every objection — that no observation could possibly count against it. What looks like robustness is in fact emptiness: a claim that explains everything explains nothing.
i.
Conspiracy"The lack of evidence for the conspiracy is proof of how thorough the cover-up has been." — Every absence becomes confirmation; nothing could refute the theory.
ii.
Pop psychology"Deep down, everyone is selfish. Even acts of apparent kindness are secretly self-serving." — The "secretly" clause renders the thesis immune to all evidence to the contrary.
iii.
Markets"My investment strategy is sound; the market is simply irrational and hasn't caught up yet." — Indefinitely deferred, no result counts as falsification.
❦ ❦ ❦
Part IV
Errors of Language
When the words themselves do the misleading
Natural language is generous: it permits ambiguity, double meanings, vagueness of boundary, and the silent reshaping of an opponent's view. Each of those generosities is also an opening for fallacy. Here the defect is not in the inference but in the way words are made to do more, or different, work than the reasoner is willing to acknowledge.
21
Equivocation
A word that shifts its meaning mid-argument
A key term is used in two different senses across the argument, producing what looks like a valid inference. Detection requires substituting precise paraphrases for each instance of the term; the apparent connection usually evaporates.
i.
Classical"The end of a thing is its perfection. Death is the end of life. Therefore death is the perfection of life." — "End" shifts from "purpose" to "termination."
ii.
Civic"We have a right to free speech, so my company can't fire me for what I say." — "Right" shifts from constitutional protection against state action to private-employment expectation.
iii.
Statistical"The average family has 2.3 children. Therefore some family must have a fractional child." — "Average" shifts from arithmetic mean to typical case.
22
Amphiboly
Structural ambiguity — the sentence that can be parsed two ways
Where equivocation exploits the multiple meanings of a single word, amphiboly exploits the multiple readings produced by syntax. Sloppy grammar, dangling modifiers, ambiguous reference: each can be cynically deployed, or innocently produced and then opportunistically exploited.
i.
Headline"Police shoot man with knife." — Who held the knife? The grammar leaves it open.
ii.
Oracle"If Croesus crosses the river, a great empire will fall." — The Delphic priestess did not specify whose empire.
iii.
Contract"The agent will be paid a commission on every sale he makes from the showroom." — Does "from the showroom" modify "sale" or "makes"? Whole lawsuits have hinged on such things.
23
Accent
Meaning shifted by emphasis, italicization, or selective stress
The same words carry different implications depending on which are stressed. By choosing where to place the emphasis — or in print, what to bold or italicize — one can subtly reframe a claim while pretending only to report it.
i.
Insinuation"I didn't say she stole the money." vs. "I didn't say she stole the money." vs. "I didn't say she stole the money." — Five identical words; five different accusations.
ii.
Advertising"Now with 30% LESS FAT*" — *than our previous, especially fatty, version. The shouted number does the persuading; the asterisk does the truth-telling.
iii.
Reportage"The official has so far denied any wrongdoing." — "So far" carries the insinuation; remove it and the sentence is exonerating.
24
Quoting Out of Context
Contextomy — the quote that lies by truncation
A passage is extracted and presented in a way that materially distorts what the speaker meant. The words themselves may be exact; the surrounding qualifications, the rhetorical purpose, or the addressee that made the original meaning clear are stripped away.
i.
Film criticismA reviewer writes: "If you enjoy lazy plotting and shoddy effects, this film is a delight." The poster: "...this film is a delight." — The Reviewer.
ii.
Political"You did not build that" — lifted from a speech about infrastructure ("Somebody invested in roads and bridges. If you've got a business — you did not build that") and repackaged as a denial of entrepreneurial achievement.
iii.
Scripture"There is no God." — Psalm 14:1, taken alone. The verse in full: "The fool hath said in his heart, There is no God."
25
Straw Man
Refuting a misrepresented version of the opponent's view
Rather than engage the actual argument an opponent makes, the reasoner constructs a weaker, more extreme, or simply different version — one easier to demolish — and demolishes that. The opposite virtue is the steel man: restating an opponent's view in its strongest possible form before responding.
i.
Policy"You want gun-safety legislation? So you want to disarm law-abiding citizens and let criminals run wild." — A modest proposal is recast as confiscation plus anarchy.
ii.
Education"You're saying we should teach critical thinking? You just want to fill kids' heads with cynicism toward authority." — A pedagogy is reframed as a vendetta.
iii.
Environment"Environmentalists want us to live in caves and abandon all modern technology." — The actual range of environmental positions, vast and diverse, is reduced to a parody.
26
The Sorites Fallacy
The fallacy of the heap — vagueness exploited to absurdity
Because the line between two states (bald vs. not bald, rich vs. not rich) is vague, the reasoner concludes there is no real distinction at all. The lack of a sharp boundary is treated as the absence of any boundary, and small incremental changes are claimed to never produce qualitative difference.
i.
Classical"One grain of sand is not a heap. Adding one grain to a non-heap doesn't make it a heap. Therefore no number of grains constitutes a heap."
ii.
Ethics"A small bribe isn't really corruption. And a slightly larger one is barely different from a small one. So there's no clear line where it becomes corruption at all."
iii.
Diet"One cookie won't ruin my diet. And one more is hardly different from one. So I can have as many as I like."
❦ ❦ ❦
Part V
Errors of Cause & Composition
When sequence is mistaken for cause, and parts for wholes
Causal reasoning is among the most error-prone of all our inferential habits. We are pattern-detectors by evolutionary necessity, and our pattern-detectors fire generously even when nothing real is there. The fallacies in this part — with their close cousins, the fallacies of composition and division — collect the most common ways in which we conjure a connection that is not present, or transfer a property from level to level where it does not survive the journey.
27
Post Hoc Ergo Propter Hoc
"After this, therefore because of this"
Temporal succession is mistaken for causal connection. Because B followed A, A is held to have caused B. Yet the world is dense with coincidences, and time alone provides only the most preliminary clue to causation.
i.
Folk medicine"I took the herbal tea, and the next morning my cold was gone. The tea cured me." — Colds resolve on their own; the tea may have done nothing.
ii.
Sports"I wore my lucky socks and we won. The socks did it." — The team's performance, the opponent's mistakes, and the referee's calls are all silently demoted to coincidence.
iii.
Politics"The new tax cut passed in January, and the economy grew that year. The cut caused the growth." — Possibly — but global conditions, monetary policy, and many other forces also acted that year.
28
False Cause — Correlation Mistaken for Causation
Cum hoc ergo propter hoc — "with this, therefore because of this"
When two variables move together, there are at least four live possibilities: A causes B, B causes A (reverse causation), a third variable C causes both (confounding), or the coincidence is mere chance. To pick the first without ruling out the others is to mistake correlation for causation.
i.
Reverse causation"Teams that pass the ball more often lose more games — passing leads to losing." — In fact, losing teams pass more because they are trailing and must catch up.
ii.
Confounding"Children who eat breakfast get better grades. So breakfast improves school performance." — Likely there is a third factor — engaged, organized parenting — that produces both the breakfast and the grades.
iii.
Pure coincidence"Ice-cream sales correlate strongly with drownings. Ice cream causes drowning." — Both are caused by summer heat. The correlation is real and the causal claim is absurd.
29
The Slippery Slope
The unargued chain of catastrophe
A first step is opposed not for its own consequences but for an undefended chain of further steps it is alleged to make inevitable. The fallacy lies in asserting the chain rather than demonstrating each link. Many slopes are not slippery; many chains have natural stopping points.
i.
Civic"If we let students wear hats in class, soon they'll be wearing pyjamas; soon there will be no standards at all." — Headwear and pyjamas are governed by distinct norms; the slide is asserted, not shown.
ii.
Legal"If we permit medical use of this drug, we'll eventually have to permit recreational use, and then every drug will be legal." — Distinct legal frameworks, accumulated evidence, and political checks all interpose.
iii.
Personal"If I have one drink at the party, I'll have ten, and then I'll lose my job, my marriage, and my life." — Possibly true for someone with a specific history; an unargued generalization otherwise.
30
The Fallacy of Composition
Inferring from part to whole
A property is true of the parts of a thing; the reasoner concludes that it is therefore true of the whole. Sometimes the inference holds (each brick is heavy; the wall is heavy). Often it does not (each atom in the cat is colorless; the cat is not colorless).
i.
Economics"If one spectator stands up at the stadium, she sees better. Therefore if all spectators stand up, all will see better." — The paradigmatic fallacy of composition: what works for one fails when generalized.
ii.
Saving"If one household saves more, it grows wealthier. Therefore if all households save more, the nation grows wealthier." — Keynes's paradox of thrift: aggregate saving collapses demand and shrinks the very income from which the savings come.
iii.
Sport"Every player on this team is excellent — so the team must be excellent." — Teams depend on chemistry, role-fit, and coaching; a roster of all-stars can underperform a balanced unit.
31
The Fallacy of Division
Inferring from whole to part
The mirror image of composition. A property of the whole is attributed to its parts, where there is no warrant. The university is excellent; therefore every department in it must be excellent. Sometimes the inference holds, often it does not.
i.
Academic"This university is world-renowned, so the philosophy department must be world-renowned." — Departments vary widely within strong institutions.
ii.
Statistics"The average household income in this town is $80,000, so the family next door earns about that." — The mean of a distribution tells us little about any particular member.
iii.
Reputation"The orchestra is one of the finest in Europe — so any musician you pluck from it will be an exceptional soloist." — Ensemble strength rests on cohesion; soloists are a different test.
❦ ❦ ❦
Part VI
Errors of Induction
When evidence is misweighed, samples mislead, and pattern is mistaken for signal
The final part collects the fallacies most likely to corrupt our reasoning about evidence. The age of data has made these errors at once more dangerous and more diagnosable: dangerous because the appearance of statistical rigour now lends false confidence; diagnosable because we have inherited from probability theory and modern epidemiology a vocabulary precise enough to name the slippages.
32
Weak Analogy
A comparison stretched past its joint
Analogies are powerful instruments of thought, but they hold only insofar as the compared cases share the features that matter for the conclusion. A weak analogy seizes on superficial similarity to import a verdict from a case whose relevant properties differ.
i.
Governance"Running a country is just like running a business — so we need a businessman in charge." — The features that matter most (coercive authority, public goods, electoral accountability) are precisely the ones business and state do not share.
ii.
Biology"The brain is just a computer — so once we build a powerful enough one, it will think like us." — The analogy illuminates and misleads in roughly equal measure; embodiment, plasticity, and chemistry differ deeply.
iii.
Education"Teaching children is like programming computers — just install the right modules." — The metaphor breaks where it matters: motivation, embodiment, social context.
33
Hasty Generalization
From too few, too quickly
A universal or near-universal claim is drawn from a sample too small, or too unrepresentative, to support it. The most reliable signature is the size of the leap from "I have encountered N cases" to "X is generally true."
i.
Travel"I went to Paris and the two waiters I met were rude. Parisians are rude." — A sample of two from a city of millions.
ii.
Medical"My grandfather smoked until 95 and never got cancer. Clearly the link is overstated." — One survivorship case against decades of epidemiological data.
iii.
Hiring"The last three candidates from that university disappointed us — we'll stop recruiting there." — Three out of how many graduates? The selection mechanism, not the school, may be at fault.
34
Sweeping Generalization
The fallacy of accident — a rule applied where its exceptions live
The mirror of hasty generalization: a sound general rule is mechanically applied to a case that lies outside its intended scope. The rule may be true on average, in the typical case, or under default conditions — but the case at hand is precisely one of the exceptions.
i.
Civic"Citizens have the right to free assembly. Therefore the prisoners should be allowed to gather wherever they wish." — The general right does not survive the carceral context.
ii.
Medical"Exercise is good for the heart. Therefore this patient, three days after open-heart surgery, should run a marathon." — A general truth misapplied to a contraindicated case.
iii.
Ethical"Honesty is a virtue. Therefore I'll tell my friend exactly what I think of his late wife at the funeral." — A virtue does not exhaust its application; context constrains.
35
Biased Sample
Selection that prejudges the result
Even a large sample can be useless if its members were not drawn in a way that represents the underlying population. The most common offenders are self-selection, survivorship, and availability — we count those who chose to be counted, those who survived to be counted, or those most visible to us.
i.
Self-selection"95% of our online poll respondents oppose the policy." — The people who took the trouble to respond are exactly those most aroused against the policy.
ii.
Survivorship"Vintage products were built to last — just look how many old radios still work." — The radios that broke long ago are not in the sample; you only see the survivors.
iii.
Availability"Plane crashes are everywhere in the news; flying must be terrifying." — News covers what is rare and dramatic; daily car-crash deaths vastly exceed annual air fatalities.
36
The Anecdotal Fallacy
The vivid story trumping the dull aggregate
A specific, emotionally resonant case is allowed to override large-sample statistical evidence. The single story is concrete and memorable; the statistics are abstract and forgettable. The fallacy is in letting the vividness do the epistemic work.
i.
Vaccination"My cousin's child got autism right after his vaccinations — I don't care what the studies say." — The studies aggregate millions of cases; the cousin's child is one.
ii.
Consumer"This brand has terrible reliability scores, but my uncle's lasted twenty years — I'm getting one." — The uncle is an n of one against an n of thousands.
iii.
Career"Everyone says I should finish my degree, but my friend dropped out and made millions — so will I." — The visible success masks the invisible majority of drop-outs whose paths did not converge on millions.
37
The Gambler's Fallacy
The belief that independent events "balance out"
In a sequence of genuinely independent trials, the outcome of the next trial is unaffected by past outcomes. The reasoner believes a "run" of one outcome makes the opposite outcome "due." A fair coin that has just landed heads ten times is no more likely to land tails next than it ever was: the probability remains one-half.
i.
Roulette"Red has come up seven times in a row — black must be due." — The wheel has no memory; the eighth spin is its own event. Monte Carlo, 1913: 26 reds in succession ruined many a player who reasoned exactly this way.
ii.
Family planning"We've had four boys — the next one is bound to be a girl." — Each conception's odds are essentially independent.
iii.
Investing"This stock has been down ten days running — it has to bounce." — In an efficient market, prior moves do not entitle one to a reversal.
38
The Hot-Hand Fallacy
The gambler's fallacy run in reverse
Past streaks are taken not as making a reversal more likely (the gambler's error) but as making continuation more likely. A run of successes is taken as evidence of an underlying "hot" state. The fallacy lies in overestimating how much of the apparent streak is signal and how much is the run of clusters that random sequences naturally produce.
i.
Basketball"He's made his last four shots — pass it to him, he's on fire." — The classic case; decades of research showed the effect is weaker than fans believe, though not zero.
ii.
Trading"My last three trades all worked — I'm in a groove. I'll size up the next one." — Three trials say little about underlying skill; sizing up exposes the reasoner to a streak of regression.
iii.
Hiring"Our last four hires from that school were stars — let's lock in a pipeline." — Four cases is a thin reed; the next four may well regress.
39
Base Rate Neglect
Forgetting how rare the underlying condition is
When evaluating the probability of a specific case, the reasoner attends to the vivid features of that case (its similarity, its narrative fit) while ignoring the base rate at which such cases occur in the larger population. Even a very accurate test for a very rare condition will produce more false positives than true ones in raw numbers.
i.
Medical screening"This test for a rare disease is 99% accurate, and I tested positive — I'm 99% likely to have it." — If the disease afflicts one in ten thousand, most positives, even from a 99% test, will be false positives. Bayes' theorem does the calculation.
ii.
Profession"Lin is shy, bookish, and quiet — probably a librarian, not a salesperson." — The base rates differ: salespeople vastly outnumber librarians; even a moderate match probability favours the larger category.
iii.
Security"This terrorism-detection algorithm is 99.9% accurate; we should screen every passenger." — Against a base rate of perhaps one terrorist per ten million travelers, the false-positive flood would dwarf the true positives.
40
The Texas Sharpshooter
Pattern painted on after the bullets land
A cluster is identified in random data, then a hypothesis is constructed to "explain" that cluster, as if the cluster's location had been predicted in advance. The reasoner ignores the vast space of possible patterns that could have emerged but didn't; one of them was bound to. The name comes from the marksman who fires at a barn and then paints the bull's-eye around the densest concentration of holes.
i.
Epidemiology"There's an unusual cluster of a rare cancer in this neighborhood — it must be the power lines." — In a country with thousands of neighborhoods, clusters of rare events will arise by chance; the cluster did not predict the explanation.
ii.
Numerology"Look at all the strange coincidences between Lincoln's and Kennedy's assassinations — the universe is sending a signal." — Two famous lives generate vast numbers of comparable details; some coincidences are inevitable.
iii.
Finance"Our backtest shows this trading strategy would have returned 40% per year for the last decade." — If thousands of strategies are tested on historical data, some will appear miraculous purely by chance. Out-of-sample they collapse.
❦ ❦ ❦
The Practitioner's Checklist
Ten Questions to Ask of Any Argument
- Is the inference valid in form? — If we replaced the subject matter with abstract letters, would the structure still hold? Look first for affirming the consequent, denying the antecedent, undistributed middles.
- Are the premises about the same thing as the conclusion? — If premises and conclusion are about different topics, you are likely seeing a red herring, an ad hominem, or some other defect of relevance.
- Does any premise quietly contain the conclusion? — Try rewriting each premise without using the contested term. If you can't, the argument may be begging the question.
- Are the options as restricted as the speaker claims? — Whenever you hear "either / or," sketch the actual landscape of possibilities. False dilemmas dissolve under the question: what's the third option?
- Does the same word mean the same thing each time it appears? — Substitute precise paraphrases. Equivocation hides in vague or polysemic terms doing different work in different sentences.
- Has the opponent's view been represented at its strongest? — If the version being attacked seems easy to refute, ask whether anyone actually holds that view, or only the steel-manned cousin of it.
- For any causal claim: what are the alternative explanations? — Reverse causation, confounding, chance, selection. Until they are addressed, the causal hypothesis is one candidate among many.
- Is the sample representative, and large enough? — Ask who is in the sample, how they got there, and who is missing. Self-selection and survivorship are silent everywhere.
- What is the base rate? — Before judging how likely a specific case is, ask how common that kind of case is in the population. Most diagnostic intuitions improve dramatically when the base rate is made explicit.
- What would change my mind? — If the answer is "nothing — any evidence to the contrary can be reinterpreted," the claim is self-sealing and should be treated with suspicion regardless of its content.
Set in Fraunces & Newsreader·Forty Specimens·Read it Twice